AWS sells out cloud capacity through 2027 as AI demand grows

When AWS says its cloud capacity is largely “sold out,” it means customers have already reserved or committed to use most of the computing resources AWS expects to have available through 2027. This is not the same as every AWS service being completely unavailable. AWS is still keeping some capacity flexible for other customers.

The clearest example is AWS’s statement that its 2026 capacity is already sold out. Jassy also said AWS is holding back part of its 2027 capacity for on-demand customers and services such as Bedrock and EC2. This creates room for customers that need computing without making long-term commitments.

The situation shows how strongly AI demand is pressuring cloud infrastructure. AWS has commitments extending into 2028, while other major cloud providers also face constraints. Amazon says it plans about US$220 billion in capital expenditure in 2026 and could invest even more to expand capacity.

AWS is looking several years ahead: its capacity is largely sold out through 2027, and some customer commitments already extend into 2028. Jassy specifically said AWS’s 2026 capacity is sold out. That gives a clear measure of how far AI-related demand has moved beyond the short term.

Amazon previously said it would spend about US$220 billion in capital expenditure during 2026. This spending is intended to support the company’s broader infrastructure needs, including the capacity AWS provides to customers. The figure represents planned investment, not a guarantee that every dollar will go directly toward one service or one type of chip.

Jassy said Amazon “could invest a lot more” and that doing so “wouldn’t be a bad use of capital.” The implication is that Amazon sees continued demand for cloud infrastructure. Still, adding capacity takes time, so spending more may not immediately remove shortages through 2027.

The article identifies AWS customers as the parties reserving capacity, but it does not name particular companies. Their commitments reflect demand for cloud computing, especially as AI workloads require substantial infrastructure. AWS’s 2026 capacity is already sold out, and most 2027 capacity is also committed.

AWS is nevertheless holding back some 2027 capacity for on-demand customers and services including Bedrock and Elastic Compute Cloud, or EC2. On-demand capacity can be used by customers that need resources without making the same long-range commitments. It also keeps important AWS services available for changing or immediate workloads.

This balancing act matters because long-term reservations provide planning certainty, while on-demand capacity provides flexibility. AWS is trying to support both types of use while supply remains tight. The article does not identify the individual customers or specify how much 2027 capacity is being held back.

Amazon EC2, or Elastic Compute Cloud, is a service that lets customers rent virtual computing resources from AWS. They can choose computing power, storage, and networking for applications. Bedrock is a managed AWS service that helps developers use foundation models and build generative AI applications through cloud-based tools.

For example, a company could use Bedrock to access a model for text generation, then use EC2 instances for supporting software, data processing, or a customized machine-learning workload. The key mechanism is shared cloud infrastructure: AWS supplies servers, chips, networking, and related services, while customers access them remotely.

The article names both services as examples of where AWS is preserving some 2027 capacity. That matters because AI applications can create heavy demand for computing. EC2 and Bedrock therefore compete for infrastructure within the broader AWS system, even though they serve different layers of an AI application.

AI systems can require much more computing than many conventional software applications. Training models involves processing enormous datasets repeatedly, while serving AI responses requires computing each time users submit requests. As more businesses experiment with or deploy AI, demand can rise faster than infrastructure can be built.

The main mechanism is a supply chain with several bottlenecks. Cloud providers need specialized chips, servers, networking equipment, data-center space, and electricity. They must install and connect these components before customers can use them. Even if Amazon is willing to spend more, construction, equipment delivery, and power availability can limit how quickly capacity grows.

The article directly links AWS’s capacity constraints to rising AI demand and says other large cloud providers face similar pressure. AWS capacity is largely sold out through 2027, with commitments into 2028. That timing shows the imbalance is expected to persist beyond a short-term spike.

When cloud capacity is scarce, companies may not receive the computing resources they want exactly when they want them. They could face delays launching AI products, limits on how quickly applications grow, or pressure to make longer-term commitments. The article does not state that AWS customers are already experiencing each of these outcomes.

A company training or running an AI system might therefore need to reserve capacity well in advance. It could also use AWS’s available on-demand capacity, adjust its workload, or consider another cloud provider. These responses depend on the customer’s needs and on what resources are available.

The broader warning is that AWS is not alone. Other large cloud providers also face constraints linked to AI demand. AWS has sold out its 2026 capacity and largely committed 2027 capacity, while Amazon says it could invest more. Capacity expansion may improve availability, but the article gives no timetable for ending the constraints.

Cloud computing is the delivery of computing resources through remote data centers. Customers access processing power, storage, networking, and software services instead of building and operating all the equipment themselves. AWS is Amazon’s cloud computing division, providing this infrastructure to customers through services such as EC2 and Bedrock.

A cloud provider must run physical data centers filled with servers and specialized chips. Those machines process customer workloads and connect them through high-speed networks. Facilities also need cooling, reliable electricity, security, and backup systems. AI can intensify these requirements because training and serving models can demand substantial computing resources.

That physical foundation explains why capacity cannot expand instantly. AWS must obtain equipment, build or equip facilities, and supply enough power before new resources become usable. The article reports that Amazon plans about US$220 billion in 2026 capital expenditure and could invest more as AI demand continues to outpace supply.